Hello,
I used the gls function from the nlme package to run a generalized least squares
model. One of the predictor variables is a factor with 3 levels. Here is a
reproducible example:
library(nlme)
response <- c(rnorm(5,1,3), rnorm(5,6,1), rnorm(5,10,5))
foo <- data.frame(response = response,
X=rep(letters[1:3], each=5),
Y=rep(LETTERS[1:3], each=5))
m1 <- gls(response ~ X, weights = varIdent(form= ~1|Y), data=foo)
The anova command indicates that the factor X is significant:
anova(m1)
The summary command compares the mean of each level of X to the reference level,
which is 'a' in this case:
summary(m1)
Based on the summary command, I will report that levels 'b' and
'c' are greater than 'a' at the p < 0.05 level. My question
is, what test should I cite for these post hoc comparisons? Are these contrasts
a version of Tukeys, Scheffe, Fisher LSD, or something similar.
This reproducible example can also be viewed at:
http://rpubs.com/jbeaulie/12839
=================================Jake J. Beaulieu, PhD
US Environmental Protection Agency
National Risk Management Research Lab
26 W. Martin Luther King Drive
Cincinnati, OH 45268
USA
513-569-7842 (desk)
513-487-2511 (fax)
beaulieu.jake@epa.gov
[[alternative HTML version deleted]]